Feature Request: Support for Gemini, DeepSeek, and OpenAI (Multi-LLM Fallback)

[kayacihan]

First of all, thanks for this awesome project. It's a lifesaver for podcast listeners.

Currently, the server is hardcoded to use Claude (Anthropic). While Claude is great, it’s not always the most accessible option for everyone due to pricing or regional availability.

I’d like to suggest adding support for other LLMs, specifically Gemini 1.5 Flash (which has a very generous free tier) and DeepSeek (which is extremely cheap).

Proposed Idea: It would be amazing if we could implement a fallback system in the ad detection logic:

Primary: Use Gemini (via Google AI Studio or OpenAI-compatible API).

Fallback 1: DeepSeek V3 (very cost-effective).

Fallback 2: GPT-4o-mini.

Final Fallback: Claude.

Since most of these providers now support the OpenAI SDK format, we could probably achieve this by just making the base_url, model_name, and api_key configurable in the .env file.

Why this helps:

Lower Cost: Users can run this almost for free using Gemini's free tier.

Reliability: If one API is down or hits a rate limit, the server can still process the podcast.

Flexibility: Users can choose their preferred "brain" for ad detection.

[Basti-Fantasti]

This would be awesome. Especially when listening to many and long podcasts (sometimes 4h+) the Claude API eats up a lot of money in a short time

[ttlequals0]

I have created a fork of this project that supports using different LLM providers, but not fallback ones.
https://github.com/ttlequals0/minuspod